Predicting future cognitive decline with hyperbolic stochastic coding.

Area of Science:

• Neuroimaging
• Computational Geometry
• Machine Learning

Background:

• Hyperbolic geometry models brain surfaces but yields high-dimensional features, limiting statistical power in cognitive decline prediction.
• Existing surface-based brain morphology analysis methods face challenges with limited subject data.

Purpose of the Study:

• To propose a novel framework, hyperbolic stochastic coding (HSC), for effective feature dimension reduction in brain imaging.
• To enhance the statistical power of cognitive decline prediction using brain morphology analysis.

Main Methods:

• Computing diffeomorphic maps to a hyperbolic parameter space for shape feature extraction.
• Utilizing farthest point sampling with breadth-first search for ring-shaped patch generation.
• Applying stochastic coordinate coding and max-pooling for feature dimension reduction.

Main Results:

• The proposed HSC framework demonstrates superior classification accuracy on Alzheimer's disease progression datasets.
• Preliminary experiments indicate improved performance compared to existing methods.

Conclusions:

• HSC offers a promising tool for surface-based brain imaging research.
• The algorithm may serve as a diagnostic and prognostic indicator for individualized treatment strategies.